Multiplexing, Switching and Routing -...
Embed Size (px)
Transcript of Multiplexing, Switching and Routing -...

Leiden University. The university to discover. 1
Multiplexing���”From one channel to multiple channels”
How to share one medium while facilitating multiple channels of communication:
Frequency Division and Time Division Multiplexing

Leiden University. The university to discover. 2
Frequency Division Multiplexing (FDM alla FSK)
Σ subcarrier fsc1
subcarrier fsc2
demodulator fsc1
band pass filter fscn
band pass filter fsc2
band pass filter fsc1
Receiver fc
Transmitter fc
subcarrier fscn
demodulator fsc2
demodulator fscn

Leiden University. The university to discover. 3
Standards # voice channel
bandwidth spectrum US/AT&T ITU_T
12 48 kHz 60-108 kHz Group Group
60 240 kHz 312-552 kHz Super Group Super Group
300 1,23 MHz 812-2044 kHz Super Group Master Group
600 2,52 MHz 564-3084 kHz Master Group Master Group
900 3,87 MHz 8,52-12.39 MHz Master Group Super Master Group
3600 16,98 MHz 0.56-17,55 MHz Jumbo Group Jumbo Group
10800 57,44 MHz 3,12-60,57 MHz Jumbo Group Multiplexed
Jumbo Group Multiplexed

Leiden University. The university to discover.
Example: ADSL ADSL Asymmetric Digital Subscriber Line
Originally for Video-on-Demand: less control going up – lots of image going down
è Very similar to internet usage !!!!!!!
Multiple “regular” phone connections at the same time on which QAM (Quadratic Amplitude Modulation) is implemented
4

Leiden University. The university to discover.
• Reserve lowest 25 kHz for Voice (POTS, Plain Old Telephone Service) 25 instead of 5 to prevent cross talk between voice&data
• Facilitate two bands: small upstream / big downstream
• Use FDM within upstream and downstream band
5

Leiden University. The university to discover.
ADSL 4 kHz channels
6

Leiden University. The university to discover.
ADSL standards
7

Leiden University. The university to discover. 8
Time Division Multiplexing(TDM)
Buffer
Buffer Modem
Buffer
Buffer
Buffer
Buffer
1 2 n n1 2 …………… n 1 2…………… ……………
channel
Frame
Synchronous TDM: not synchronous but frames are fixed and slots are always filled

Leiden University. The university to discover. 9
How is framing implemented
0101 ….. control channel
Added digit framing
Pulse Stuffing
Frequency > Σ (freq. of the sources)
additional bits are added at fixed position in the frame

Leiden University. The university to discover. 10
Relationship with data link framing
F1f2A1F2C1A2d1C2d1d2.................
for
F1f1f1d1d1d1C1A1F1 in characters flag control
information FCS
address

Leiden University. The university to discover. 11
Example: Telephony DS-1 transmission format
Voice è PCM (8000 samples per second, 8-bit) TDM-frame = 24 (channels) x 8 bits +1 (frame bit) = 193 bits Data rate: 8000 x 193 = 1.544 Mbps DATA only 23 out of 24 channels used, 24th channel has special SYNC BYTE per channel 1 bit for user/system data
è 7 x 23 x 8000 = 56 x 23 kbps = 56 kbps p. channel

Leiden University. The university to discover.
Standards Telephony US/JAPAN ITU-T
# channels Mbps Level #channels Mbps
DS-1 24 1.544 1 30 2.048
DS-1C 48 3.152 2 120 8.448
DS-2 96 6.312 3 480 34.368
DS-3 672 44.736 4 1920 139.264
DS-4 4032 274.176 5 7680 565.148
12

Leiden University. The university to discover. 13
Network SWITCHING Next to multiplexing: switching is required to realize multi to multi connections Especially needed in Wide Area Networks (WAN) Also present in Local Area Networks (LAN) or in multi processors architectures.

Leiden University. The university to discover.
Circuit Switching
14
1. A dedicated path between two end stations is realized or channel (TDM/FDM)
2. Data is being transmitted (Switches don’t inspect data) 3. Path is broken up

Leiden University. The university to discover.
Circuit switching
IMPORTANT CHARACTERISTIC:
BLOCKING VS NON-BLOCKING
15
Connection cannot be realized because all paths are occupied
Connection can always be realized and at any time

Leiden University. The university to discover.
Space Division Switching: ���non-blocking Crossbar Switch
16
Very costly: N2 switches

Leiden University. The university to discover.
Multistage Networks Use many small crossbar switches and connect them wisely.
Blocking can occur!!!!!
17
8
6
3
1
5 cannot be connected to 2

Leiden University. The university to discover.
Omega Networks���(based on Perfect Shuffles)
18
2logN + 1 stages: non blocking with O(NlogN switches)

Leiden University. The university to discover.
Variants of PS networks���Cube Network
19

Leiden University. The university to discover.
Butterfly Network
20

Leiden University. The university to discover.
Fat Tree Network
21

Leiden University. The university to discover. 22
Time Division Switching Do not confuse with TDM !!!
Bus
Bandwidth of Bus > Σ indiv. bandw.
Then non-blocking!

Leiden University. The university to discover.
Routing in Circuit Switched Networks
• Alternate Routing è Each switching node has its own routing table • Fixed Alternate Routing
Routing tables do not change • Dynamic Alternate Routing
Depending on time (e.g. time of the day) routing tables will change
• Adaptive Routing Central Controller gets status of all switches and gives routing updates to all switches
23
First choice Second choice
A to B Via switch i Via switch j
A to C Via switch j Via switch k

Leiden University. The university to discover.
Packet Switching
• Datagram Every packet is routed independently è As a consequence packets can arrive out of order
• Virtual Circuit (Wormhole) Before communication is initiated a Call-Request packet is sent on the network, which fixates a virtual path between sender and receiver. è Packets arrive in order, but it not as flexible as datagram
24
Data is sent by packets (usually < 1000 octets), Every switching nodes has buffers

Leiden University. The university to discover.
Summarizing
25
Circuit Switch Virtual Circuit Datagram
ACK signal
Call request
Call accept

Leiden University. The university to discover.
Different Combinations
• External Virtual Circuit & Internal Virtual Circuit
• External Virtual Circuit & Internal Datagram
• External Datagram & Internal Virtual Circuit
• External Datagram & Internal Datagram
Which one makes sense?
26

Leiden University. The university to discover.
Routing Trade Offs
Correctness Simplicity
Robustness Stability
Fairness Optimality
Efficiency
27

Leiden University. The university to discover.
Routing for Packet Switched Networks
Like circuit switching can we differentiate between: Fixed Routing and Alternate Routing è No difference between datagram and virtual circuit
Random Routing Every node chooses randomly outgoing link, based on some prob. Distribution, e.g. Pi = Ri / ΣRj
with Ri data rate possible on link i.
28

Leiden University. The university to discover.
Flooding (very inefficient, very reliable) Every node puts incoming packet on every outgoing link, except the incoming link è Exponential growth 1. Every node logs all the packets If packets arrives a second time: discard 2. Every packet, contains counter: hop-count If hop-count > threshold: discard
Adaptive Routing (Central vs Distributed) Every node gets network status information Ø Local, e.g. queue length of the outgoing links Ø Adjacent nodes Ø All nodes
29

Leiden University. The university to discover.
ARPANET���based on Adaptive Routing, Distr. & Adjacent Nodes
First Version: 1969 Based on Bellman-Ford Algorithm Every node i has two vectors:
Di = , Si = Every 128 ms every node exchanges delay vector with adjacent nodes. Then every node k: dkj = MiniεA [dnew
ij + dki ] and skj = i, the node i which minimizes dkj. Link delays are the queue length for that link.
Disadvantages: Link delays were not accurate Thrashing would occur
30
di1 di2
diN
si1 si2
siN
N = #nodes dij = estimated delay from node i to j sij = next node on the route i to j

Leiden University. The university to discover.
2de Generation (1979) Every node: Ø Timestamp on incoming message (arrival time) Ø Departure time recorded Ø If pos. ACK is received: delay = (dept. time – arrival time)
Every 10 sec: every node computes the average delay per link If delay is different: FLOODING is used to inform all the other nodes Every node gets status of the whole network!!!!!!!!! Dijkstra’s shortest path algorithm is used to compute new routing table
31

Leiden University. The university to discover.
3de Generation (1987) When load is heavy: Observed delay under old routing ≠ delay under new routing è Oscillation effects è Instead of BEST route: a “good” route Smoothening of link costs (delays) Every 10 seconds: 1. (Queuing theory) ρ = 2(s-t)/(s-2t), with
32
ρ=link utilization t = measured delay s = service time
2. U(n+1) = 0.5ρ(n+1) = 0.5U(n), U(n) average utilization 3. New delays are computed based on U(n), terrestrial: 1 Hop for U(n)<0.5, 2 Hops for U(n)>0.8 sattelite: 2 Hops for U(n)<0.8
Otherwise the same as 2-de generation